LANL-2010
Fluid Mechanics and Turbulence in theWind-Turbine Array Boundary Layer
Charles Meneveau,
Mechanical Engineering & CEAFM,
Johns Hopkins University
Mechanical Engineering
Collaboration with
• Prof. Raúl B. Cal (Portland State Univ.) - exp
• Prof. Luciano Castillo (RPI) - exp
• Marc Calaf (JHU & EPFL) - LES
• José Lebrón-Torres (RPI) - exp
• Dr. Hyung-Suk Kang (JHU) - exp
• Mr. Max Gibson (Portland State Univ.) - exp
• Prof. Johan Meyers (Univ. Leuven) - LES
Funding: NSF CBET-0730922 (Energy for Sustainability)
Simulations: NCAR allocation (NSF)
Mechanical Engineering
Shell's Rock River windfarm in Carbon County, Wyoming, USASource: http://www.the-eic.com/News/Archive/2005/May/Article503.htm
Land-based HAWT Horns Rev HAWTCopyright ELSAM/AS
Arrays are getting big:
When L >> 10 H (H: height of PBL),
starts to approach “fully developed”
The windturbine-array boundary layer (WTABL)
Photo by Uni-Fly A/S (Wind turbine maintenance company)
The windturbine-array boundary layer (WTABL)
Long downstream effects:
Horns Rev wind farm:
Christiansen & Hasager: “Wakeeffects of large wind farms identifiedfrom satellite synthetic aperture radar(SAR)”, Remote Sensing Env. 98, 251(2005)
Modelling and measurements of wakes in large wind farmsBarthelemie, Rathmann, Frandsen, Hansen et al…J. Physics Conf. Series 75 (2007), 012049
Power extraction at Horns Rev wind farm:
Long downstream effects:
10H ! 10km : 10km7D
=10km
0.56km= 18
?
Importance of wake fluid dynamics and turbulence
Elkinton, Manwell & McGowan (2006)Offshore Wind Farm Layout Optimization (OWFLO) Project: Preliminary Results AIAA paper AIAA-2006-998
Total cost optimization for various layouts.
PARK model for wake velocity reduction (wake loss):
Fluid dynamics and details of wake structure very important all the way to the economic analysis
Sensitivity of cost to variouscomponents included in
their cost analysis:
U(x) =U0 1! a1
1+"(x / D)#$%
&'(
2)
*++
,
-..
Entrainment parameters depend on turbulence structure…
Sample of works on modelling and measurements of wind turbine wakes and arrays:
• Lissaman PBS. Energy effectiveness of arbitrary arrays of wind turbines. In: AIAAPaper 79-0114. p. 1-7, 1979
• Voutsinas SG, Rados KG, Zervos A. On the analysis of wake effects in wind parks.Journal of Wind Eng Ind Aerodyn 14, 204-219, 1990.
• S Frandsen: wind speed reduction… J. Wind Eng & Ind Appl 39, 1992.• Crespo, Hernandez & Frandsen: Survey of modeling methods… Wind Energy 2, 1-24,
1999.• Magnusson M, Smedman AS. Air flow behind wind turbines. J Wind Eng Ind Aerodyn
80, 169- 89, 1999.• Frandsen & Thogersen: Integrated fatigue loading for wind turbines in wind farms by
combining ambient turbulence and wakes. J Wind Engin 23, 327-339, 1999.• Ebert & Wood: Renewable Energy, detailed single wake measurements.. 1999.• Vermeer, Sorensen & Crespo: Wind turbine wake aerodynamics. Prog Aerospace Sci
39, 467-510, 2003.• Corten, Schaak & Hegberg: Turbine interaction in large offshore wind farms, Report
ECN-C-04-048, 2004.• Medici & Alfredsson: Measurement on a wind turbine wake: 3D effects and bluff body
vortex shedding. Wind Energy 9, 219-236, 2006.• Jakob Mann: Simulation of Turbulence, Gusts and Wakes for Load Calculations, Wind
EnergyProceedings of the Euromech Colloquium, 2007.• Chamorro & Porté-Agel: A wind-tunnel investigation of wind-turbine wakes: Boundary-
layer turbulence effects Wind Energy, 2009 (in press)• …..
The “fully developed” WTABL:
z
U(z) = u (x, y, z) xy
U0
x
y
Assume flow driven by imposed (unaffected) dpdx(depends on Ug, and we disregard “turning” effects)
I = V/R
Models for R so that we may evaluate flow (I) for a given driving force (V)
• Momentum theory (time averaged + “canopy average”):
0 = !1"dp#dx
+ddz
! u 'w 'xy! u "w" xy( ) + fx xy
The “fully developed” WTABL:
z
U(z) = u (x, y, z) xy
Horizontal averageof turbulent Reynolds stresses
We must include correlationsbetween mean velocity deviations
from their spatial mean(Raupach et al. Appl Mech Rev 44, 1991, Finnigan,
Annu Rev Fluid Mech 32, 2000)
u " = u ! u xy
U0
x
y
Assume flow driven by imposed (unaffected) dpdx(depends on Ug, and we disregard “turning” effects)
thrust force due to WT
0 = !1"dp#dx
+ddz
! u 'w 'xy! u "w" xy( ) + fx xy
The fully developed WTABL: momentum theory
If top of WT canopy still
falls in the “surface layer”, where
and if wakes have “diffused” so that
! u 'w 'xy(ztop ) " ! u 'w '
xy(zbottom ) +
12CT
AdiskAxy
UR2
u "w" xy ! 0
dpdxz ! 0
Sten Frandsen, J. Wind Eng & Ind
Appl 39, 1992):
Horizontally averaged variables
u*2hi ! u*
2lo +
12CT
"4sxsy
UR2
Integrate in z-direction:
The fully developed WTABL: momentum theory
u*2hi ! u*
2lo +
12CT
"4sxsy
UR2
Frandsen 1992: postulated the existence of 2 log laws with u*hi and u*lo
u xy = u*hi1!log z
z0,hi
"
#$%
&'
u xy = u* lo1!log z
z0ground
"
#$%
&'
u xy (z)
log(z)zh
The zeroth-order question:
Flow resistance as effective roughness length
S. Frandsen, 1992:
u xy = u*hi1!log z
z0,hi
"
#$%
&'
This is of the form “I = V/R”
I = velocity (output),
V=u* is applied driving force, and
R=resistance:
If z0,hi is known, we can relate u*
and mean velocity u(z)
Note: for Geostrophic forcing, u* must
Further be related to geostrophic wind
(straight-forward if z0,hi is known)
R =1!log z
z0,hi
"
#$%
&'"
#$
%
&'
(1
The fully developed WTABL concept and effective roughness: sample application
Keith et al. “The influence of large-scale wind power on climate” PNAS (2004)
Barrie & Kirk-Davidoff: “Weather response to management of large Wind turbine array”, Atmos. Chem. Phys. Discuss. 9, 2917–2931, 2009
(under review for ACP)
Use z0,hi ~ 0.8 m - assumes
rough-wall log-law (MO)
Grid-spacings 100’s of km,
first vertical point ~ 80m
The zeroth-order question:
Flow resistance as effective roughness length
u*2hi ! u*
2lo +
12CT
"4sxsy
UR2
S. Frandsen 1992, 2006:
UR = u*hi1!log zh
z0,hi
"
#$%
&'3 unknowns: z0,hi , UR , u* lo
Knowns: u*hi , z0,ground , CT , sx , sy
u*hi1!log zh
z0,hi
"
#$%
&'= u* lo
1!log zh
z0,ground
"
#$%
&'
Solve for effectiveroughness:
z0,hi = zh exp !" #CT
8sxsy+
"ln(zh / z0,ground )
$
%&'
()*
+,,
-
.//
!1/2$
%&&
'
())
u xy = u*hi1!log z
z0,hi
"
#$%
&'
u xy = u* lo1!log z
z0ground
"
#$%
&'
u xy (z)
log(z)zh
Another important question: Fluxes of kinetic energy
c
d
a
b
VV (1! 2a)
For single wind turbine, extracted power = difference in front and back fluxes of kinetic energy
Another important question: Fluxes of kinetic energy
For multiple (∞ ) wind turbines in fully developed WTABL, extracted power = must be brought to wind turbine
by vertical fluxes of kinetic energy:
U0
x
y
d 12 u xy
2
dt= !"turb ! "canop !
ddz
u 'w 'xyu xy + u "w" xy u xy( ) ! u xy
1#dp$dx
! PT (z)
Objectives:
• Measure profiles of horizontally averaged momentum fluxes & mean velocity for various wind turbine arrangements and conditions (neutral for now)
• Verify existence of double log-law structure, check Frandsen formula for z0,hi, develop improvements?
• Quantify fluxes of kinetic energy: correct magnitude? Turbulence or mean flow?
• Challenges: requires statistics at entire 3-D volume surrounding a turbine elementary volume, including in transverse direction
• LES: fully developed, but contains many modeling assumptions
• Wind tunnel: more realistic, but not fully developed and exp. uncertainties
Large Eddy Simulation
• Closure problems:
- need a model for (Lagrangian dynamic Smag. model)
- need a model for wall stress (log-law wall function)
- need a model for
Subgrid scalestress tensor
! ij = uiu j! " "ui "u j
!wi
fTi
! !ui!t
+ !uk! !ui!xk
= "1#!!p!xi
"!!xk
$ ik "1#! !p!xi
+ fTi
• LES equations (formally filtered Navier-Stokes eqns. - here: neutral):
Wind-turbineforce model
Imposed PGforcing of flow
Actuator disk modeling of turbines in LES
fTx = !12CT
11! a
U"#$
%&'2 (Ayz
(V= !
12
)CTU2 (Ayz
(V
CT = 0.75! a " 0.25# $CT = 1.33
Jimenez et al., J. Phys. Conf. Ser. 75 (2007) simulatedsingle turbine in LES using dynamic Smag. model
U(t) = (1! ")U(t ! dt) + "Udisk (t)
fTx = !12CTUref
2 "Ayz
"V, CT = 0.75
They used fixed reference (undisturbed) velocity:
Here we use disk-averaged and time-averaged velocity, but local at the disk (see Meyers & Meneveau 2010, 48th AIAA conf., paper)
Simulations setup:
H = 1000 !1500m, Lx = "H ! 2"H , Ly = "H(Nx # Ny # Nz ) = 128 #128 #128
• Code: horizontal pseudo-spectral (periodic B.C.), vertical: centered 2nd order FD(Moeng 1984, Albertson & Parlange 1999, Porté-Agel et al. 2000, Bou-Zeid et al. 2005)
• Top surface: zero stress, zero w
• Bottom surface B.C.: Zero w + Wall stress: Standard wall function relating wall stress to first grid-point velocity
• Scale-dependent dynamic Lagrangian model (no adjustable parameters)
• Calaf, Meneveau & Meyers, (Phys Fluids 2010, 22)
Simulations results:
Instantaneous stream-wise velocity contours:
top-viewside-view
front-view
Simulations results:
Instantaneous stream-wise velocity contours:
Simulations results:
Instantaneous power output (mean over all WT and ±rms):
Simulations results:
Instantaneous power output (along rows and columns):
Comparison of wake profiles, regular Smagorinsy (with wall damping) and dynamic model:
Dynamic Smagorinsky coefficient: increases in wake region, while decreases near wall
Simulations results:
cs2 1/2
Mean velocity profile:
Simulations results: horizontally averaged profiles
u xy
Log-law without WTSame slope,higher-z intercept(is z0,hi)
Lower slope (u*,lo)
Mean Reynolds stress profile:
Conclusions: Momentum fluxes carried by cross-stream heterogeneity are significant
Simulations results: horizontally averaged profiles
Spatial “canopy stress” profiles,! "u w ' xy ! u "w" xy
Suite of LES cases:
Suite of LES cases:
measure z0,hi from intercept
Improvement to Frandsenʼs model:
z0,hi = zh 1+ D2zh
!"#
$%&
'
exp ()CT
8* 2sxsy+ ln zh
z0,ground
1( D2zh
!"#
$%&
'+
,--
.
/00
!
"##
$
%&&
(2+
,
---
.
/
000
(1/2!
"
###
$
%
&&&
where ' =1w
*
1+ 1w* , and 1w
* =1T
*u*zh , eddy viscosity due to wake
Comparison of LES results with models:
Triangles: Lettau formula Asterisks: Frandsen (2006) formula
Circles: improved Frandsen modelCalaf, Meneveau & Meyers, (Phys. Fluids 2010, 22)
z0,hi = zh 1+ D2zh
!"#
$%&
'
exp ()CT
8* 2sxsy+ ln zh
z0,ground
1( D2zh
!"#
$%&
'+
,--
.
/00
!
"##
$
%&&
(2+
,
---
.
/
000
(1/2!
"
###
$
%
&&&
where ' =1w
*
1+ 1w* , and 1w
* =1T
*u*zh , eddy viscosity due to wake
z0,hi = zh exp !" #CT
8sxsy+
"ln(zh / z0,ground )
$
%&'
()*
+,,
-
.//
!1/2$
%&&
'
())
Using the roughness model for array optimization - find s-opt:
Using the roughness model for array optimization - find s-opt:
Using the roughness model for array optimization - find s-opt:
Using the roughness model for array optimization - find s-opt:
• Lease of land, yearly payout per wind turbine is around US$ 5,000 for present typical spacings of 500x500m (e.g. http://www.windustry.org/how-much-do-farmers-get-paid- to-host-wind-turbines).
• So over a 20-year lease: pL ≈ 0.4 US$/m2.
• Land purchase e.g. in Texas ~ US$ 1,000 per acre (e.g. http://recenter.tamu.edu/data/agp),
or costL ≈ 0.25 US$/m2
• Representative cost of a wind-turbine US$3.5x106 for a 2MW-rated wind turbine. With D= 70m: costT/A ≈ 700 US$/m2
(e.g. http://www.windustry.org/how-much-do-wind-turbines-cost).
• The corresponding parameter α is then roughly
α = (costT/A)/costL ~ 1,700 to 3,000
Using the roughness model for array optimization - find s-opt:
At common s ~ 7D, 10-20% suboptimal possible reason for “array underperformance” ?
Meyers & Meneveau, 2010 (preprint, submitted to Wind Energy)
α ~ 2,000
Wind-tunnel measurements: mechanics of vertical KE entrainment??
Contractionsection
CR=25:1
Corrsin Wind Tunnel (1966): Test Section (1.2m×0.9m)
Flow
Rough surface
1x
2x
0.7 m
D
2.9 m
D = 12 cmModel wind turbine
Flow
Strakes
1x
2xD
2.9 m
Active gridKang et al. (JFM 2003)
Wind-tunnel measurements: mechanics of vertical KE entrainment??
Wind-tunnel measurements
Flow
Strakes
1x
2xD
2.9 m
Strakes
Wind-tunnel measurements
Flow
Strakes
1x
2x
hh=12cm
2.9 m
Wind turbine models
•Turbine Models–Scaled down 850 times from typicalreal life length scales (real diametersof 100 m scaled down to 12 cm).
• Rotors–Made from G28 galvanized sheet metal–Twisted 1.1 degrees per cm, from 15o at the root to 10o at the tip–Tip speed ratio, λ = Vtip/Uhub is 5–Rotate at 4800 RPM
• Tower– Height of 12 cm– Constructed using rapid prototyping
D=12cm
Wind-tunnel measurements
Flow
Strakes
1x
2xD
2.9 m
optical sensorfor phase-lock and
Ω rpm measurements
Hot-wire inlet profiles
u* = 0.48 m/s
x
y
1.4 m
u =u*!ln yy0
=0.480.4
ln y3.5 "10#4
TI∞ ~ 6%
TSI System with:
• Double pulse Nd:YAG laser(120mJ/pulse)
– Laser sheet thickness of 1.2 mm
– Time between pulses of 50 ms
– Optical sensor external trigger for phaselock measurements
• Two high resolution cross/autocorrelation digital CCD cameras with
– a frame rate of 16 frames/sec.
– Interrogation area of 20 cm by 20 cm
Mirror
20 cm
Laser Sheet
20 cm
3rd Row ofwind turbines
Phase-lockSensorFlow
Stereo-PIV system
PIV data planes:
3 cm
6 cm
18cm
18cm
Top view:
Statistics:
2000 vector maps for each front plane12000 samples each back plane (6 phase-locked cases)
Wind-tunnel measurements
Velocity maps:
3.7D
3 D
Mean streamwise velocity
Velocity maps:
3.7D
Mean transverse velocity
Wake angular momentum
Velocity maps:
3.7D
3 D
Mean vertical velocity
Velocity maps:
(negative) Reynolds shear stress
Cross-stream maps:
(negative) Reynolds shear stressStreamwise velocity
Measured induction factor:
a =121! Uback
U front
"
#$%
&'( 0.087
Horizontally (canopy) averaged profiles:
“copy” - assume periodic+ linear interpolation
Mean velocity(a) in front(b) in back(c) overall
Horizontally (canopy) averaged profiles:
Horizontally averaged Reynolds stresses:
u*L ! 0.1 ! 0.32 m/s u*H ! 0.28 ! 0.53 m/s
Cal et al.: J. RenewableAnd Sustainable Energy 2, 2010
Horizontally (canopy) averaged profiles:
u = u*L1!ln yu*L
"#$%
&'(+ 5.5 ) *U +#
$%&'(
!U + " 2.2
u*L ! 0.32 m/s
Power extraction:
P =12!CP
"4D2UR
3
CP! ideal = 4a(1! a)2 = 4 "0.087(1! 0.087)2 # 0.29
P =12!1.2 !0.29 ! "
4!0.122 !6.243 # 0.47 W
Direct torque measurements:
Torque
Back View
Strain Gage 1
Strain Gage 2
DC Motor
HousingHousing
Model generator(DC Motor)
Ball Bearings
Side View
Shaft
DC motor as a generator
Vary R
V
Thanks to Duane Dennis (High-Schooljunior at Baltimore Polytech Inst.)for his help with these experiments
Direct torque measurements:
Therefore at 4800 rpmPreal = T meas !" = 0.34 W
# CP-real =Preal
12 $%R
2UR3 & 0.21
Measured torque
Telec =VI!
Tmec
Ohm
72% of ideal 0.47 W for given a
About 50% of real-life wind turbines
Small except in back
Horizontally averaged profiles - kinetic energy terms:
d 12 u xz
2
dt= !"turb ! "canop !
ddy
u 'v 'xzu xz + u "v " xz u xz( ) ! u xz
1#dp$dx
! PT (y)
Horizontally averaged profiles - kinetic energy terms:
d 12 u xz
2
dt= !"turb ! "canop !
ddy
u 'v 'xzu xz + u "v " xz u xz( ) ! u xz
1#dp$dx
! PT (y)
u 'v 'xzu xz (ytop ) ! u 'v '
xzu xz (ybottom ) " 1.4 W / m2
(kg / m3) #Pturb! flux = 1.4$APturb! flux = 1.4 %1.2 % (3% 0.12)(7 % 0.12)Pturb! flux = 0.51 W
Analysis consistent with view thatkinetic energy extracted by turbine(0.34W) is delivered vertically byturbulence fluxes (0.51W)(rest goes into dissipation, etc…)
Quadrant and spectral analysis of vertical KE entrainment:
“Quadrant Analysis”:contributions from u’ > or < 0 v’> or < 0Fi = ! u 'v ' Qi u (y)
u '
v 'Q1Q2
Q4Q3
From: Gibson, Cal & Meneveau (2010)
Main conclusion:Q2 (ejections) + Q4 (sweeps) dominant
Quadrant and spectral analysis of vertical KE entrainment:
“Co-spectral Analysis”F = ! u(y) "uv (kx , y)# dkx
From: Gibson, Cal & Meneveau (2010)
Main conclusion:Large scales dominate entrainment process (~ π/k ~ 10-30 cm = 1-3 D)
LES: Horizontally averaged profiles - kinetic energy terms
d 12 u xy
2
dt= !"turb ! "canop !
ddz
u 'w 'xyu xy + u "w" xy u xy( ) ! u xy
1#dp$dx
! PT (z)
• LES and data so far confirms that overall, boundary layer momentum theory(friction velocity, thickness length-scales, momentum loss, etc..) givesvaluable insight and parameterizations of dynamics.
• LES confirms existence of 2 log-laws below and above WT region
• We have extended Frandsen model for effective roughness height, withimproved agreement with LES results.
• The notion of the fully developed WTABL is a useful theoreticalnotion, otherwise mix of BL + wake, scale separation, 3-D complex flowcomplicates statistical treatment and data analysis.
• How much momentum transfer caused by mean velocity dispersive stresses?we found them small in experimentwe found them large in LES
• Energetics and profiles show that kinetic energy is delivered to turbine mostlyby turbulence (both LES and experiments confirm this).
• Expand experiments with model rotors with higher induction factor, vary λ...
• Model for effective roughness height (R) can be used to design optimal powerextraction per cost (terrain and # of WT)
Conclusions: